Issue forecast looks to predict the total amount of time it takes for an issue to be completed, capturing the initial move to in progress to the final date it was closed. In order to form a prediction for open issues, we build a model based off the customers/individuals historical behavior, leveraging data from both the issue tracking and source code platforms. To highlight a few key features used in the model, we consider the age of the issue (time issue was created to started), how many times the issue was edited since being created, day of the week issue was started, the type of issue (bug, feature, task, etc). Furthermore, we use Natural Language Processing around the issue's description and title, grouping issues together based off their respective topics, and then feed these topics into the model.
Example of Issue forecasting and Sprint Risk
In collaboration with issue based features, source code features we currently incorporate are the assignee's historic cycle time, and the assignee's teams historic merge time (time from PR created to merged). The model is constantly evolving as more features get added and we continue to address various clientele behavior.
In order to form as accurate predictions as possible, the model takes into account all issues that spent time in an "in progress state". Therefore, issues that simply were transitioned from the backlog to closed for instance, are not taken into consideration for the model.
When interpreting forecast in the app, the estimated time is assumed from the issue's start date, and does not account for how much time the issue has already been in progress for. In an effort to better assist planning and encompass the issue's true lifecycle, the forecast accounts for calendar days, not work days. For instance if an issue is estimated to have a forecast of 24 hours; that signifies 24 hours from when the developer starts the issue. This is a true 24 hours, not 24 working hours.
https://www.youtube.com/watch?v=8GQdszUX6ys
Pinpoint Founder and CEO, Jeff Haynie walks through Pinpoint's machine learning behavior for Sprint Forecasting and Risk. Learn how Pinpoint and our customers are using the stand-up view to get proactive vs reactive during the sprint process, and ultimately get more consistent with their overall deliverability.